Need advice (or maybe a mentor) for building an ML pipeline for sales forecasting

projects
1

Hey everyone :waving_hand:

I’m currently working on a project where I need to forecast meat sales per product (article) — sounds simple, but it’s getting a bit messy.

Right now, I have an Excel sheet with 4 product articles. I train a separate model for each one (so 4 models in total).

But soon there might be 150–200 articles, and I’ll need to automate everything — so that each model updates every 2 weeks or once a month and gives daily forecasts.

Basically, I need to figure out:

  1. how to build a scalable pipeline (so I don’t have to train hundreds of models manually),

  2. and later, how to handle MLOps (deployment, updates, retraining, etc.).

I’m pretty new to this side of ML engineering, and of course, time is tight (don’t even ask how I ended up in this situation :sweat_smile:).

I’m looking for someone who could maybe point me in the right direction, share some experience, or even mentor or collaborate on this.

Any advice is appreciated — where to look for information, what tools or frameworks might fit, or how to structure things so it scales nicely.

If needed, I can share a small example of the data and part of my code.

Thanks in advance to anyone who takes the time to help :folded_hands:

2

Before building anything, do you actually need 200 separate models? This is the biggest decision. Right now you’re thinking 4 products = 4 models, so 200 products = 200 models. But this approach doesn’t scale well not just technically, but conceptually.

However, if your products are genuinely different, then separate models might make sense. Also, before going into MLOps, it’s not a bad idea to run everything manually and then automate each part. start with one product see what it takes there is no one right answer you can build one and keep iterating. This way you’ll have traceability to what and why if something is breaking, and can also focus on one step at a time.

1 Like